Magnetic Domains and Domain Walls

Ferromagnetism left this chapter a promissory note. It showed that a demagnetized iron crystal is not disordered but districted — locally saturated domains whose directions cancel — and it priced the arrangement with a slab-model budget: a 1 cm cube of iron saves half a joule of field energy by splitting into a few thousand domains, each wall costing about a millijoule per square metre. But it deliberately treated the wall as a line item, a price tag with no anatomy. Why does a wall cost what it costs? Why is it roughly two hundred atoms thick rather than one? And why did the budget's answer — domains a few microns wide — depend on the sample being a centimetre across? This article opens the wall. Inside it, the magnetization does not flip; it rotates, smoothly, through every intermediate direction, across a ribbon tens of nanometres wide. The width is a negotiated settlement between two energies pulling opposite ways: exchange, the quantum spring that hates any angle between neighboring spins and therefore wants the rotation spread over as many atoms as possible, and magnetocrystalline anisotropy, the crystal's preference for certain easy axes, which taxes every spin caught pointing between them and therefore wants the rotation over with quickly. The compromise gives the wall its width $\delta = \pi\sqrt{A/K}$ and its energy $\sigma_w = 4\sqrt{AK}$ — and once $\sigma_w$ is in hand, the domain pattern itself follows, including the square-root-of-size law the slab budget promised. The same competition, run at small scale, produces one of the chapter's most consequential results: below about ten nanometres a particle cannot afford a wall at all and stays single-domain — the physics on which every hard disk ever built depends.

Key takeaways

A domain wall is a finite ribbon, not a surface: the magnetization rotates continuously from one domain's direction to the next over $\delta = \pi\sqrt{A/K} \approx 64$ nm in iron — about 220 lattice spacings.
The width is a compromise: exchange stiffness $A$ (quantum, quoted forward from Exchange Interaction and Ferromagnetic Ordering, $A \approx 2\times10^{-11}$ J/m for iron) wants slow rotation over many sites; anisotropy $K$ (crystalline easy axes, $K_1 \approx 4.8\times10^4$ J/m³ for iron) wants the detour through hard directions kept short.
The wall's price is $\sigma_w = 4\sqrt{AK} \approx 3.9$ mJ/m² for iron — the number Ferromagnetism's domain budget spent on credit, here derived.
Balancing wall cost against field-energy savings gives the equilibrium domain width $d = \sqrt{6\sigma_w L/\mu_0 M_s^2} \propto \sqrt{L}$ : micron domains in centimetre crystals, and finer domains — not coarser — as samples shrink.
Below $R_{sd} = 9\sigma_w/\mu_0M_s^2 \approx 10$ nm a particle cannot repay a wall's cost and stays single-domain; a few nanometres smaller still and thermal energy flips the whole moment at once — superparamagnetism, the size floor of magnetic recording.
KEY IDEA: exchange and anisotropy negotiate the wall ( $\delta = \pi\sqrt{A/K}$ , $\sigma_w = 4\sqrt{AK}$ ); the wall and the stray field negotiate the domain pattern ( $d \propto \sqrt{L}$ ); and when the sample gets small enough, the negotiation collapses to a single domain.

The core insight

Domain walls are finite-width ribbons whose width and energy are set by the competition between exchange (favoring gradual rotation) and magnetocrystalline anisotropy (favoring alignment with easy axes); the wall energy in turn sets the equilibrium domain size, the single-domain limit, and the superparamagnetic floor beneath all magnetic storage.

Where this lives today

Domain walls stopped being a microscopy curiosity and became a device component. In racetrack memory (IBM's Stuart Parkin, and a research field since), bits are domains punched into a magnetic nanowire and the walls between them are shifted lock-step by spin-polarized current pulses — storage with no moving atoms, only moving walls, gliding at hundreds of metres per second. Magnetic sensors in every car ABS ring and laptop lid use the field-driven motion of Néel walls in permalloy films. And the wall's exotic cousins — skyrmions, chiral bubbles stabilized by the Dzyaloshinskii–Moriya interaction in thin films lacking inversion symmetry — are 2026's densest candidates for wall-based logic, small enough that a single skyrmion approaches the superparamagnetic arithmetic this article ran. Meanwhile the blocking size stopped being a nuisance and became a tool: MRI contrast agents and magnetic-hyperthermia cancer therapies use deliberately superparamagnetic iron-oxide nanoparticles — big moment, zero remanence, so they respond strongly in the scanner but never clump in the bloodstream. The same $KV/k_BT$ exponential that limits hard disks makes them safe.

Origins · 1 minFelix Bloch and how this was discovered

Felix Bloch

1905–1983 · Zurich, Leipzig, Stanford

Bloch put the wall in the wall's name. Heisenberg's new quantum mechanics had explained why spins align (exchange); his student Bloch asked what the aligned state's excitations and textures look like. His 1930 spin-wave theory gave the $T^{3/2}$ law for how magnetization melts; his 1932 habilitation analyzed the boundary between oppositely magnetized regions and showed it must be a smooth, many-atom rotation — thickness $\propto\sqrt{A/K}$ — not an atomic flip. The "Bloch wall" is that calculation. The rest of the résumé is absurdly deep: Bloch electrons in periodic potentials (his 1928 thesis, the foundation of band theory), the Bloch equations of nuclear magnetic resonance, and the 1946 discovery of NMR in condensed matter that shared the 1952 Nobel Prize with Purcell — the physics inside every MRI magnet. Driven out of Leipzig in 1933, he became Stanford's first theoretical physics professor and, in 1954, the first Director-General of CERN. This article is his wall, priced and put to work.

History

1907

— Weiss postulates spontaneous magnetization and domains, with no mechanism for either boundary or alignment

1919

— Barkhausen hears domain avalanches: magnetization proceeds in discrete wall jumps, not smoothly

1931

— Bitter decorates domain boundaries with magnetic colloid: the first direct images of walls

1932

— Bloch shows the wall is a smooth many-atom rotation with width set by exchange against anisotropy

1935

— Landau and Lifshitz found micromagnetics: flux-closure domains and the equilibrium pattern as energy minimization

1955

— Néel predicts the in-plane wall of thin films, completing the Bloch–Néel dichotomy

2008

— Parkin's racetrack memory proposal makes the moving domain wall itself the storage element

The physical picture

Stand inside the crystal at a wall and look along the line of atoms crossing it. On the far left every atomic moment points up; on the far right, down. If the reversal happened between two adjacent atoms — one up, its neighbor down — the exchange interaction, which pays an energy penalty growing as the square of the angle between neighbors, would be charged its absolute maximum at that one bond. Exchange is the strongest interaction in the problem (it is what holds the Curie temperature at 1043 K, four orders of magnitude above anything dipolar — the verdict Ferromagnetism computed), so an atomically sharp wall is the single most expensive way to reverse. The crystal's escape is to spread the reversal: rotate each moment by a small angle relative to its neighbor, so no single bond pays much. If the rotation spans $N$ atoms, each of the $N$ bonds carries an angle $\pi/N$ , each pays energy $\propto (\pi/N)^2$ , and the total exchange bill falls as $1/N$ . Left to exchange alone, the wall would fatten without limit — a mile-wide wall would cost almost nothing. Something must push back, and it is the crystal lattice itself. In iron the moments prefer to lie along the cube edges — the easy axes; pointing along a face diagonal or body diagonal costs extra energy. This magnetocrystalline anisotropy is quantified by a constant $K$ (energy per volume). Every atom inside the wall is, by construction, caught pointing somewhere between two easy directions, so every atom in the wall is paying the anisotropy tax — and the bill grows linearly with the wall's thickness. Exchange pulls the wall wide; anisotropy squeezes it thin. The equilibrium width is the geometric-mean compromise between the two, and the wall's energy per unit area, $\sigma_w$ , is what both taxes together cost at the optimum.

Figure

Inside a 180-degree domain wall - exchange spreads the rotation while anisotropy taxes every tilted spin

The two energies, named honestly

Both contestants deserve their credentials before they fight. The exchange side enters as the exchange stiffness $A$ , the continuum packaging of the quantum interaction whose microscopic derivation is owned by Exchange Interaction and Ferromagnetic Ordering, the chapter's closer. When the magnetization direction $\mathbf{m} = \mathbf{M}/M_s$ varies in space, the exchange energy density is

u_{\rm ex} = A\,|\nabla\mathbf{m}|^2 \qquad (1)

where: $A$ — exchange stiffness, joules per metre; $\mathbf{m}$ — the unit vector along the local magnetization; $|\nabla\mathbf{m}|^2$ — the summed squared gradients of its components. Equation (1) is the continuum limit of "each misaligned neighbor bond costs energy growing as its angle squared." The stiffness inherits its scale from the ordering temperature: exchange must be worth roughly $k_BT_C$ per atom, spread over a lattice spacing $a$ , so $A \sim k_BT_C/2a \approx 2\times10^{-11}$ J/m for iron ( $T_C = 1043$ K, $a \approx 0.29$ nm) — the loan we take from the exchange article and repay by using it well. The anisotropy side is introduced here phenomenologically, the way a magnetician meets it in a datasheet. Crystals are not isotropic to their own magnetization: spin–orbit coupling ties the moments weakly to the lattice, making some directions cheap (easy axes) and others expensive. For a uniaxial material the leading term is

u_{\rm anis} = K\sin^2\theta \qquad (2)

where: $K$ — the first anisotropy constant, joules per cubic metre; $\theta$ — the angle between $\mathbf{M}$ and the easy axis. For iron (cubic, but the uniaxial form captures the wall physics) $K_1 \approx 4.8\times10^4$ J/m³. Note the scales: anisotropy is tiny per atom — $K a^3 \sim 10^{-6}$ eV against exchange's $10^{-1}$ eV. That thousand-fold mismatch is exactly why walls are hundreds of atoms wide rather than one: the strong interaction sets the penalty for speed, the weak one the penalty for delay, and the compromise lands far on the side of the strong one.

Derivation AThe wall from a spin chain - width and energy from two competing taxes

The goal is the pair of results this chapter has been spending on credit: the wall width $\delta = \pi\sqrt{A/K}$ and the wall energy per area $\sigma_w = 4\sqrt{AK}$ . The strategy is to run the competition twice — first as a back-of-envelope over a discrete chain of $N$ spins, which exposes why each tax scales as it does, then properly, as a one-dimensional variational problem whose first integral does all the work. The watch-out is the boundary condition: the wall profile is pinned at $\theta = 0$ and $\theta = \pi$ only at infinity, and it is exactly this that forces the local balance between the two energy densities.

Step 1. The discrete estimate. Let the rotation from up to down be shared equally among $N$ sites of spacing $a$ : each bond carries angle $\pi/N$ . Exchange charges each bond $\sim JS^2(\pi/N)^2/...$ — in stiffness language, energy per wall area $\sim A(\pi/N a)^2 \times Na = A\pi^2/(Na)$ . The exchange bill falls as $1/N$ .

Step 2. Anisotropy charges every one of the $N$ sites for being off-axis, on average half the maximum: energy per area $\sim K\,Na$ . The anisotropy bill rises as $N$ .

Step 3. Minimize the sum $E(N) = A\pi^2/(Na) + KNa$ over $N$ : setting $dE/dN = 0$ gives $N^* a = \pi\sqrt{A/K}$ — the two bills are equal at the optimum, and the width is the geometric mean of the strong and weak scales. For iron, $N^* a = \pi\sqrt{2\times10^{-11}/4.8\times10^4} = 64$ nm, about 220 lattice constants. Why the geometric mean: exchange alone sets no length (it only wants "wider"), anisotropy alone sets no length (it only wants "thinner"); a length exists only because both are present, and $\sqrt{A/K}$ is the unique combination with units of metres.

Step 4. Now the honest continuum version. With $\theta(x)$ the rotation angle across the wall, the energy per unit area is the functional

\sigma[\theta] = \int_{-\infty}^{\infty}\left[A\left(\frac{d\theta}{dx}\right)^2 + K\sin^2\theta\right]dx

with $\theta(-\infty) = 0$ , $\theta(+\infty) = \pi$ . The integrand has no explicit $x$ -dependence, so the Euler–Lagrange equation has a first integral: $A\,\theta'^2 - K\sin^2\theta = \text{const}$ , and the constant is zero because both terms vanish at infinity. Hence, locally and everywhere in the wall,

A\left(\frac{d\theta}{dx}\right)^2 = K\sin^2\theta

Why this matters: it is equipartition made local — at every point in the wall the exchange density equals the anisotropy density. The discrete estimate's "two equal bills" was not an accident of the optimum; it holds point by point.

Step 5. Separate and integrate: $d\theta/\sin\theta = \sqrt{K/A}\,dx$ gives $\ln\tan(\theta/2) = x/\lambda$ with $\lambda = \sqrt{A/K}$ , i.e.

\theta(x) = 2\arctan\left(e^{x/\lambda}\right)

a smooth kink that spends most of its rotation within $\pm\lambda$ of the centre. The conventional width takes the slope at the centre and extrapolates: $\delta = \pi\lambda = \pi\sqrt{A/K}$ .

Step 6. The energy. Using the first integral to convert the $x$ -integral to a $\theta$ -integral, the two equal terms give

\sigma_w = 2\int_{-\infty}^{\infty} K\sin^2\theta\,dx = 2\sqrt{AK}\int_0^{\pi}\sin\theta\,d\theta = 4\sqrt{AK}

For iron: $\sigma_w = 4\sqrt{(2\times10^{-11})(4.8\times10^4)} = 3.9$ mJ/m². Verified: profile and constants match Kittel ch. 12 and Chikazumi; the iron numbers recomputed for this edition give $\delta = 64$ nm $= 223a$ and $\sigma_w = 3.9$ mJ/m²; the equipartition first-integral trick is the same one that solves the soliton profile of the sine-Gordon equation — the Bloch wall is, structurally, a topological soliton.

One reconciliation is owed immediately. Ferromagnetism's domain budget used $\sigma_w \approx 1.5$ mJ/m² and got 5-micron domains; our derived value is 3.9 mJ/m². Both are honest. The $4\sqrt{AK}$ result is the idealized 180° Bloch wall in a perfect uniaxial crystal; measured 180°-wall energies in real iron span roughly 1–4 mJ/m² depending on crystallographic orientation, purity, and how the cubic anisotropy actually routes the rotation. The conclusions are insensitive: rerunning the slab budget with 3.9 mJ/m² moves the domain width from 5 to 8 microns — the same order, the same physics, the same verdict that centimetre iron carries micron domains.

The square-root law the budget promised

Ferromagnetism minimized its slab budget at fixed sample size $L = 1$ cm and found $N^* \approx 2000$ walls. Held there, the result looks like a number; freed, it is a scaling law — the one piece of the budget that article explicitly deferred here.

Go deeper · Derivation B · 1 minDomain width grows as the square root of sample size

Derivation BDomain width grows as the square root of sample size

The goal is the equilibrium domain width $d(L)$ for a sample of thickness $L$ , exposing its $\sqrt{L}$ dependence. The strategy recalls the slab budget of Ferromagnetism verbatim — uniform-magnetization field energy $U_0 = \mu_0M_s^2V/6$ traded against wall cost — and simply carries the minimization symbolically instead of numerically. Watch which quantities are per-sample and which are per-width: the field energy scales with the domain width $d$ , the wall count scales against it, and the compromise is the geometric mean once again.

Step 1. Recall the budget. A cube of side $L$ split into slab domains of width $d = L/N$ retains stray-field energy only within about one domain width of its top and bottom faces; the field bill is proportional to $d$ , normalized so that at $d = L$ (one domain) it returns the uniform bill $U_0 = \mu_0M_s^2L^3/6$ :

U_{\rm field}(d) \approx \frac{\mu_0M_s^2L^2}{6}\,d

Step 2. Price the walls: $N - 1 \approx L/d$ walls of area $L^2$ each,

U_{\rm wall}(d) \approx \sigma_w L^2\,\frac{L}{d}

Step 3. Minimize the total. $dU/dd = 0$ gives $\mu_0M_s^2L^2/6 = \sigma_w L^3/d^2$ , hence

d = \sqrt{\frac{6\,\sigma_w L}{\mu_0 M_s^2}} \qquad (3)

where: $d$ — equilibrium domain width; $L$ — sample dimension along the magnetization; $\sigma_w$ — wall energy per area; $M_s$ — saturation magnetization. For iron with $\sigma_w = 3.9$ mJ/m² and $L = 1$ cm: $d = \sqrt{6(3.9\times10^{-3})(0.01)/(\mu_0(1.7\times10^6)^2)} = 8.0\ \mu$ m. Why $\sqrt{L}$ and not $L$ : doubling the sample doubles the wall area per wall AND doubles the field bill, but the number of walls is free to adjust; the balance point moves only as the square root. Domains do not scale with their sample — a crystal twice as thick has domains only 1.4 times wider, i.e. relatively finer. *Verified: equation (3) with $\sigma_w = 1.5$ mJ/m² returns the 5 μm of Ferromagnetism's budget (4.98 μm recomputed), and with our derived 3.9 mJ/m² gives 8.0 μm — same order, reconciling the two price lists; the $\sqrt{\sigma L}$ form matches Kittel's domain-theory chapter and Landau–Lifshitz's original analysis.*

The law explains an experimental staple: Kerr-microscope images of thick crystals show coarse, branching domains, while thin films of the same material show fine, delicate patterns. It also warns that "domain size" is not a material constant — quoting one without the sample geometry is like quoting a terminal voltage without the load.

Bloch, Néel, and the closure escape

So far the wall rotated its magnetization in the plane of the wall — out of the page, like a corkscrew — which is the Bloch wall, the bulk solution. The choice is not aesthetic. Rotating within the wall plane keeps $\nabla\cdot\mathbf{M} = 0$ inside the material: no volume poles, no stray field, no magnetostatic surcharge on top of the exchange-plus-anisotropy bill we already paid. (The pole-counting doctrine — magnetic "charge" wherever $\mathbf{M}$ has divergence, surface poles where it meets a boundary — is the machinery of The Auxiliary Field H and Magnetic Boundary Conditions, spent here without re-derivation.) Confine the wall in a thin film, though, and the Bloch rotation betrays itself: mid-wall, the magnetization points out of the film's face, planting poles on both surfaces. For film thickness $t$ comparable to or below the wall width $\delta$ , the pole bill outgrows everything else, and the wall re-routes: the magnetization rotates within the film plane instead — a Néel wall — trading the surface poles for (cheaper, at this geometry) volume poles inside the wall. The crossover near $t \sim \delta$ is why permalloy films tens of nanometres thick, the raw material of magnetoresistive sensors and racetrack devices, carry Néel walls while bulk iron carries Bloch walls. The same pole-avoidance instinct sculpts the domain pattern near surfaces. Slab domains meeting a free surface head-on would plant alternating pole stripes there; the crystal prefers to cap the structure with small triangular closure domains, magnetized parallel to the surface, that pipe the flux around the corner and back down the neighboring domain — a magnetic circuit closed entirely inside the material, at the cost of some extra wall and (in magnetostrictive materials) some elastic strain. Landau and Lifshitz drew exactly this flux-closed pattern in 1935, in the paper that founded micromagnetics.

Figure

Bloch wall versus Neel wall, and closure domains piping flux around the surface

When the wall stops paying: single domains and the thermal floor

Equation (3) has a fine-print clause with industrial consequences. The domain pattern exists because splitting saves more field energy than the walls cost. Shrink the sample and both sides of the ledger shrink — but not at the same rate. Field energy is a volume effect ( $\propto R^3$ ); the wall is a cross-section effect ( $\propto R^2$ ). Below some radius the wall can no longer repay its own cost, and the particle stays uniformly magnetized no matter what: a single-domain particle.

Go deeper · Derivation C · 2 minThe single-domain radius and the superparamagnetic blocking size

Derivation CThe single-domain radius and the superparamagnetic blocking size

The goal is the two sizes that bracket magnetic recording: the radius $R_{sd}$ below which a particle cannot afford a domain wall, and the smaller blocking diameter below which even the single domain cannot hold its direction against heat. The strategy is two budget comparisons — wall cost versus field saving for $R_{sd}$ , anisotropy barrier versus thermal energy for the blocking size. Watch the units: everything is decided by the ratio of $\sigma_w$ (J/m²) and $\mu_0M_s^2$ (J/m³), which is a length.

Step 1. Price the split. A sphere of radius $R$ , uniformly magnetized, stores field energy $U_1 = \frac{\mu_0M_s^2}{6}\cdot\frac{4}{3}\pi R^3$ (the sphere coefficient from Field of a Magnetized Object, recalled). Splitting into two hemispherical domains with one equatorial wall saves roughly half the field energy while paying $\sigma_w\pi R^2$ for the wall.

Step 2. Set saving against cost: $\tfrac12 U_1 \gtrsim \sigma_w \pi R^2$ requires

R \gtrsim R_{sd} \approx \frac{9\,\sigma_w}{\mu_0M_s^2} = \frac{36\sqrt{AK}}{\mu_0M_s^2}

For iron: $R_{sd} = 9(3.9\times10^{-3})/(\mu_0(1.7\times10^6)^2) = 9.7$ nm. Why hard magnets have larger single-domain sizes: $R_{sd} \propto \sqrt{K}$ , so high-anisotropy materials keep single-domain grains at tens to hundreds of nanometres — exactly the regime permanent-magnet and recording technology exploits, since a grain with no wall can only reverse by coherent rotation, the most expensive route (the bound Hysteresis prosecutes).

Step 3. Now the thermal floor. A single-domain grain of volume $V$ holds its magnetization along an easy axis behind an anisotropy barrier $\Delta E = KV$ . Thermal agitation attempts the flip at frequency $f_0 \sim 10^9$ – $10^{10}$ Hz; the mean waiting time is $\tau = f_0^{-1}e^{KV/k_BT}$ . Demanding stability for years, not nanoseconds, requires the standard engineering criterion

KV \gtrsim 25\,k_BT

(25 gives $e^{25} \approx 7\times10^{10}$ attempts survived — about a decade at $f_0 = 10^9$ Hz).

Step 4. Solve for the diameter at room temperature with iron's $K = 4.8\times10^4$ J/m³: $V = 25k_B(300)/K = 2.2\times10^{-24}$ m³, i.e. a sphere of diameter

d_b = \left(\frac{6V}{\pi}\right)^{1/3} \approx 16\ \text{nm}

Below $d_b$ the grain is superparamagnetic: still ferromagnetically ordered inside (exchange never sleeps), but its net moment flips freely with temperature, remembering nothing. Verified: $R_{sd} = 9.7$ nm and $d_b = 16$ nm recomputed for this edition from $\sigma_w = 3.9$ mJ/m², $K = 4.8\times10^4$ J/m³, $M_s = 1.7\times10^6$ A/m; the $KV = 25k_BT$ criterion and attempt-frequency range match the Néel–Brown relaxation theory as presented in Blundell and in recording-industry literature. Note the two sizes nearly coincide for iron — soft materials make terrible recording media, which is why real media use high-K alloys where $R_{sd}$ and $d_b$ separate widely.

Here is the recording industry's fundamental dilemma in one paragraph. Denser storage wants smaller grains; smaller grains slide toward the superparamagnetic floor $d_b \propto (25k_BT/K)^{1/3}$ ; the only rescue is raising $K$ — but coercivity scales with $K$ too (the Stoner–Wohlfarth bound of Hysteresis), and a grain too stubborn to flip thermally is also too stubborn for the write head. The industry's answer — warm each bit toward its Curie point for the nanosecond of writing, then cool it back to stubbornness — is heat-assisted magnetic recording, and it is a direct application of this article's two budget lines plus Curie Law and Curie Temperature's collapse.

Limiting cases

Case	Result	Meaning
$K \to 0$ (no anisotropy)	$\delta \to \infty$	nothing taxes intermediate directions; the "wall" delocalizes — soft amorphous magnets have very wide, very cheap walls
$K$ large (hard magnet)	$\delta$ thin, $\sigma_w$ high, $R_{sd}$ large	NdFeB-class materials: nanometre walls, strongly pinned, single-domain grains — permanent magnetism by design
$A \to$ large (high $T_C$ )	$\delta \propto \sqrt{A}$ widens	stronger exchange spreads the rotation; wall width tracks $\sqrt{T_C}$ across materials
sample $L \to$ large	$d \propto \sqrt{L}$	domains grow with sample size, but relatively finer; "domain size" is geometry-dependent, never a pure material constant
$R < R_{sd} \approx 10$ nm (Fe)	no wall forms	single-domain particle; reversal only by coherent rotation — the hard-magnet and recording-grain regime
$KV < 25k_BT$	superparamagnetism	the net moment flips thermally; ordered inside, memoryless outside — the size floor of magnetic storage
film $t \lesssim \delta$	Bloch $\to$ Néel wall	surface-pole cost re-routes the rotation into the film plane; the sensor-film regime

Common trapA domain wall is not a defect and a superparamagnet is not a paramagnet

Two conflations to execute. First: the wall is not a grain boundary, dislocation, or any structural fault — it is a smooth rotation of $\mathbf{M}$ through perfect crystal, free to glide (its interactions with actual defects are the pinning story of Hysteresis). A wall can be parked anywhere; a grain boundary cannot move. Second: a superparamagnetic particle is not a paramagnet that happens to be small. Inside, exchange holds every spin rigidly aligned — the particle carries a giant moment of $10^4$ – $10^5\,\mu_B$ , not one. What it loses below the blocking size is only the direction memory of that collective moment. The signature difference: superparamagnets saturate in modest fields (giant moment, easy to align) but show zero coercivity (no memory) — a combination neither a true paramagnet nor a blocked ferromagnet can produce.

Worked examples

1 · Pricing iron's wall (confidence builder)

From $A = 2\times10^{-11}$ J/m and $K_1 = 4.8\times10^4$ J/m³, find the wall width and energy, and reconcile with the budget Ferromagnetism ran.

Step 1. Width: $\delta = \pi\sqrt{A/K} = \pi\sqrt{(2\times10^{-11})/(4.8\times10^4)} = \pi(6.45\times10^{-8}/\pi)\ldots$ compute the root first: $\sqrt{4.17\times10^{-16}} = 2.04\times10^{-8}$ m, so $\delta = \pi \times 2.04\times10^{-8} = 6.4\times10^{-8}$ m $=$ 64 nm, about $223$ lattice constants ( $a = 0.287$ nm).

Step 2. Energy: $\sigma_w = 4\sqrt{AK} = 4\sqrt{(2\times10^{-11})(4.8\times10^4)} = 4\sqrt{9.6\times10^{-7}} = 4(9.8\times10^{-4})$ = 3.9 mJ/m².

Step 3. Reconcile: measured 180°-wall energies in iron span 1–4 mJ/m²; Ferromagnetism's budget used 1.5. Rerun its domain width with our value: $d = \sqrt{6\sigma_wL/\mu_0M_s^2} = \sqrt{6(3.9\times10^{-3})(0.01)/(1.257\times10^{-6}\times2.89\times10^{12})}$ = 8.0 μm versus its 5 μm. Same order; the micron conclusion is insensitive to the factor-of-2.6 spread in $\sigma_w$ because $d \propto \sqrt{\sigma_w}$ .

2 · Designing a recording grain (exam level)

A recording alloy has $M_s = 5\times10^5$ A/m, $K = 2\times10^5$ J/m³, $A = 1\times10^{-11}$ J/m. Find the wall energy, the single-domain radius, the room-temperature blocking diameter for 10-year stability, and comment on the design window.

Step 1. Wall energy: $\sigma_w = 4\sqrt{AK} = 4\sqrt{(10^{-11})(2\times10^5)} = 4(1.41\times10^{-3})$ = 5.7 mJ/m².

Step 2. Single-domain radius: $R_{sd} = 9\sigma_w/(\mu_0M_s^2) = 9(5.7\times10^{-3})/(1.257\times10^{-6}\times2.5\times10^{11}) = 5.1\times10^{-8}$ m = 51 nm — much larger than iron's 10 nm because $M_s$ is lower (smaller field bill to save) and $K$ higher (pricier wall but even pricier field alternative scaled down).

Step 3. Blocking volume: $V = 25k_BT/K = 25(1.381\times10^{-23})(300)/(2\times10^5) = 5.2\times10^{-25}$ m³, so $d_b = (6V/\pi)^{1/3} = (9.9\times10^{-25})^{1/3} \approx 1.0\times10^{-8}$ m = 10 nm.

Step 4. The window: grains between 10 nm (thermal floor) and 51 nm (single-domain ceiling) are stable, wall-free, and switch as single units — a wide, comfortable design band, which is exactly why media engineers choose high- $K$ , moderate- $M_s$ alloys. Iron's window (16 nm floor vs 19 nm ceiling) barely exists: soft magnets cannot store bits.

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §6.4 (ferromagnetism and domains) · Jackson, Classical Electrodynamics, 3rd ed., §5.8 · Purcell & Morin, Electricity and Magnetism, 3rd ed., ch. 11 · Zangwill, Modern Electrodynamics, §13.5 · Kittel, Introduction to Solid State Physics, 8th ed., ch. 12 (the wall and domain-theory chapter this article tracks) · Chikazumi, Physics of Ferromagnetism, 2nd ed. · Blundell, Magnetism in Condensed Matter, ch. 6.

Provenance. Equations: total micromagnetic energy and exchange continuum form follow Kittel (Ch. 12) and Chikazumi; Bloch-wall results $\delta=\pi\sqrt{A/K}$ , $\sigma_{\mathrm{w}}=4\sqrt{AK}$ derived in full and verifiable against Kittel/Chikazumi. Demagnetizing-factor relation $N_x+N_y+N_z=1$ (SI) and sphere value $1/3$ are standard (Jackson §5.8, Zangwill §13.5). Kittel domain-size scaling $D\sim\sqrt{\sigma L/\mu_0 M_s^2}$ from Landau–Lifshitz closure-domain theory. Material numbers (iron: $A\sim2\times10^{-11}$ J/m, $K\sim5\times10^4$ J/m $^3$ , $\mu_0 M_s^2\sim10^6$ J/m $^3$ , wall width tens of nm) are standard textbook values; a verifier should recompute $\delta$ and $\sigma_{\mathrm{w}}$ from these and confirm $\sim60$ nm and $\sim10^{-3}$ J/m $^2$ . Conceptual treatment cross-checks Griffiths §6.4 and Purcell Ch. 11. Addendum for this edition: all numbers recomputed with $K_1 = 4.8\times10^4$ J/m³ — $\delta = 64$ nm $= 223a$ , $\sigma_w = 3.9$ mJ/m², $R_{sd} = 9.7$ nm, $d_b = 16$ nm, $d(1\,\text{cm}) = 8.0$ μm — and reconciled explicitly with the $\sigma_w = 1.5$ mJ/m², $d = 5$ μm budget of Ferromagnetism (measured 180°-wall energies span 1–4 mJ/m²; conclusions order-insensitive).

Builds on: Ferromagnetism · Magnetization · The Auxiliary Field H · Magnetic Dipole Moment of Atoms

Leads to: Magnetic Susceptibility · Linear and Nonlinear Magnetic Media · Field of a Magnetized Object

Is the magnetization weaker inside a domain wall?

Essentially no — and this is worth internalizing. Exchange is so strong that the magnitude $|\mathbf{M}| = M_s$ is locked everywhere below the Curie temperature; the wall only rotates the direction. Micromagnetics treats $\mathbf{m} = \mathbf{M}/M_s$ as a unit vector for exactly this reason. The energy scales confirm it: reducing $|\mathbf{M}|$ costs exchange energy comparable to $k_BT_C$ per atom ( $\sim 0.1$ eV), while rotating direction against anisotropy costs only $\sim 10^{-6}$ eV per atom. The crystal always buys the cheap option.

Where exactly does the wall sit? Can it be anywhere in the crystal?

In a perfect crystal, anywhere — the wall energy is position-independent, so the wall is free to glide, and an arbitrarily small field would sweep it through the sample (perfect softness). Real crystals have defects — inclusions, grain boundaries, strain fields — that locally cheapen or obstruct the wall, creating an energy landscape with valleys (pinning sites). Walls park in valleys. That landscape, its reversible bowing and irreversible jumps, is the entire subject of Hysteresis; here we only note that a mobile wall plus a rough landscape is what turns this article's statics into that article's memory.

Why is the anisotropy constant K so much smaller than exchange, and where does K come from physically?

Exchange is Coulomb energy filtered through the Pauli principle — electron-volt currency. Anisotropy comes from spin–orbit coupling: the spin senses the lattice only indirectly, through the orbital motion's weak relativistic link, so the directional preference is a high-order correction — microvolts to millivolts per atom. That hierarchy (strong but direction-blind exchange, weak but direction-aware anisotropy) is precisely why the wall is hundreds of atoms wide: the geometric mean $\sqrt{A/K}$ sits far above the lattice constant. Materials with heavy atoms (stronger spin–orbit) like Nd₂Fe₁₄B or SmCo₅ have larger $K$ and proportionally thinner, more expensive walls.

The domain-width formula says d grows with sample size. Does a huge iron beam have centimetre domains?

No — the $\sqrt{L}$ law holds for a single crystal with clean slab geometry. Structural steel is polycrystalline with grains of microns to millimetres, and domain patterns live within grains (walls cross grain boundaries only with difficulty; the boundary is a pinning wall of its own). Moreover, thick samples develop branched, hierarchical closure patterns rather than simple slabs — the scaling survives in spirit (coarser structures deeper inside, finer near surfaces) but the exponent is modified. Take equation (3) as the physics of the trade-off, quantitative in clean geometries, qualitative beyond.

If closure domains close the flux internally, why does a magnetized iron bar have any external field at all?

Because a magnetized bar is one where the applied field has deliberately destroyed the closure equilibrium: favorable domains were grown at their neighbors' expense (and pinning holds the conquest — Hysteresis's remanence). The zero-external-field, closure-capped state is the demagnetized equilibrium. The whole art of making a permanent magnet is preventing the return to it: high anisotropy, single-domain grains, and pinning so strong that the closure configuration, though energetically preferred, is kinetically unreachable. Equilibrium thermodynamics says every magnet wants to forget; the microstructure makes remembering metastable for geological times.

What actually happens when a wall meets the superparamagnetic limit - do walls fluctuate thermally too?

They do. A wall segment sitting in a pinning valley of depth comparable to $k_BT$ hops valleys thermally — this is magnetic after-effect or "creep": magnetization drifting logarithmically for seconds to years after a field change, audible as the slow tail of Barkhausen activity (the effect Ferromagnetism introduced). At the single-grain scale the analogue is the Néel–Brown flip of the whole moment. Both are the same Arrhenius physics, $\tau = f_0^{-1}e^{\Delta E/k_BT}$ , differing only in what coordinate crosses the barrier: a wall position, or the collective moment angle.

Do domain walls really move at hundreds of metres per second? What limits the speed?

Field- or current-driven walls in nanowires reach 100–1000 m/s, but not without drama: above a critical drive (the Walker breakdown), the wall's internal structure begins to precess — the plane of rotation itself rotates — and the mobility drops sharply. The wall is not a rigid object but a texture with internal degrees of freedom, and pushing it too hard excites them. Racetrack-memory engineering is substantially the art of postponing Walker breakdown (via shape anisotropy, synthetic antiferromagnets, or chiral interactions), which is why device papers obsess over wall type — Bloch, Néel, vortex — and not just wall presence.